CN101424219A - Spark plug partial mixture concentration flexible measurement method - Google Patents
Spark plug partial mixture concentration flexible measurement method Download PDFInfo
- Publication number
- CN101424219A CN101424219A CNA2007101500368A CN200710150036A CN101424219A CN 101424219 A CN101424219 A CN 101424219A CN A2007101500368 A CNA2007101500368 A CN A2007101500368A CN 200710150036 A CN200710150036 A CN 200710150036A CN 101424219 A CN101424219 A CN 101424219A
- Authority
- CN
- China
- Prior art keywords
- model
- spark plug
- network
- soft
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Combined Controls Of Internal Combustion Engines (AREA)
Abstract
The invention discloses a soft measuring method for concentration of local mixed gas of a spark plug, which comprises the following steps: A, obtaining a master variable, and establishing a mathematical model; and B, obtaining a secondary variable, and establishing a black box model; and carrying out off-line training for a BP neural network soft measuring model by an error correcting learning model, and using the trained soft measuring model in linear measurement. The soft measuring method has the advantages that the soft measuring method is fully suitable for engines of various ignition type products, not only can meet correlative fundamental research carried out in a laboratory, but also is convenient to be applied to actual vehicles, and achieves on-line real-time measurement. The measuring method can measure the concentration of the local mixed gas of the spark plug, and open up a novel research direction for optimized control strategies of forming quasi-homogeneous thin mixed gas by controllable injection oil; and test equipment for the measuring method is cheap, light and portable, and is easy to install on a vehicle.
Description
Technical field
The present invention relates to a kind of spark-ignition engine, particularly relate to a kind of detecting method of spark-ignition engine spark plug partial mixture concentration.
Background technique
Spark-ignition engine spark plug ion current signal is loaded with the abundant information relevant with the in-cylinder combustion process.Along with people to the understanding of spark plug ion current progressively deeply, testing circuit constantly improves and signal processing technology is increasingly mature, the application of spark plug ion current will provide a new technological approaches for combustion in IC engine and emission control.But directly also few at the method for spark plug partial mixture concentration detection, the mixture strength method of measurement that is adopted is laser inductive fluorescence method and Rayleigh scattering method (X mostly.Yet optical measuring method is the testing apparatus costliness on the one hand, and the measurement expense height can only be done fundamental research in the laboratory; Experiment can only be special optical engine with motor on the other hand, is difficult to be applied to production engine, and is poor for applicability, and can't obtain the result of online real-time measurement.
Summary of the invention
Technical problem to be solved by this invention is, overcomes the shortcoming of prior art, provides a kind of and utilizes a plurality of spark plug ion current characteristic parameters as auxiliary variable, based on the soft-measuring technique of BP neuron network, measures the method for spark plug partial mixture concentration.
The technical solution adopted in the present invention is: a kind of spark plug partial mixture concentration flexible measurement method is characterized in that: may further comprise the steps:
A. obtain leading variable, set up mathematical model:
(1) rotating speed, load and the fuel injection quantity of extraction motor; (2) gather ionic current, extract characteristic parameter Ip, S and Tms; (3) pass through the parameter that the switching mode lambda sensor detects oxygen; (4) set up the BP neural network soft sensor model;
B. obtain secondary variable, set up black-box model:
(1). obtain secondary variable Ip, S and Tms by the detection of spark plug ion current, oxygen sensor signal is obtained in the output of switching mode zirconium oxide formula lambda sensor, and linear air fuel ratio detects obtains A/F, filters the back and forms sample set; (2). set up black-box model; The error correction learning model carries out off-line training to the BP neural network soft sensor model, the soft-sensing model that trains is used for line measures.
Based on the BP neural network soft sensor model, create λ SS neural network model towards Matlab, wherein input vector P dimension is 4, by [Ip, S, Tms, O
2] form; Hidden layer contains 5 neurons, adopts tansig as transfer function; Output layer contains 1 neuron, adopts the purelin transfer function; The network training function is got trainlm, adopts the newff function to generate network model.
Set up the step of BP neural network soft sensor model: (1) loads sample data; (2) sample data is carried out standardization; (3) sample data is divided into training set and test set; (4) create neuron network; (5) network is trained; (6) network is carried out simulation training; (7) simulation result and target output are done linear regression analysis.
The invention has the beneficial effects as follows: flexible measurement method of the present invention is applicable to various Spark ignition type production engines fully; Not only can satisfy the relevant rudimentary theoretical research of in the laboratory, carrying out; And be convenient to the real vehicle application, realize online real-time measurement.This method of measurement can accurately have been measured the mixture strength of spark plug partial, and the Optimal Control Strategy that forms the quasi-Homogeneous weak mixture for controlled oil spout has been opened up new research direction; This method of measurement testing apparatus expense is cheap, and lightly portable, is easy to install with regard to car.
Description of drawings
Fig. 1 is the soft measurement block diagram of air fuel ratio of the present invention;
Fig. 2 is an AFR soft-sensing model of the present invention;
Fig. 3 is that ionic current characteristic parameter Ip of the present invention, S, Tms obtain the interface.
Embodiment
Below in conjunction with the drawings and specific embodiments the present invention is described in further detail:
As shown in Figure 1, utilize the spark plug ion current signal,, estimate flexible measurement method λ SS (the λ Soft Sensor) block diagram of spark plug partial mixture concentration based on the BP neural network soft sensor model.Its basic thought is: with characteristic parameters such as peak I p, the integral value S of spark plug ion current and signal duration Tms and switching mode oxygen sensor signal as secondary variable, (λ) takes variable as the leading factor with near the mixture strength spark plug, the characteristic of utilizing neuron network can fully approach the non-linear relation of any complexity makes up the black-box model of secondary variable to leading variable, characterizes the mapping of spark plug ion current signal to spark plug partial mixture concentration.Method is at first to gather learning sample, secondary variable Ip, S and Tms obtain with analytical system with the spark plug ion current detection of this paper independent development in the sample, the switching mode zirconium oxide formula lambda sensor output that oxygen sensor signal is equipped with by former machine, the linear air-fuel ratio detection system of study " teacher " (promptly surveying A/F) data records; Taked the teacher learning method that the BP neural network soft sensor model is carried out off-line training according to the learning rules of error correction then; At last the soft-sensing model that trains is applied among the on-line measurement.
The present invention adopts the identification modeling method, based on present extensive use and the comparatively ripe BP neural network theory of technology, creates λ SS soft-sensing model towards Matlab.
As shown in Figure 2, λ SS neural network model adopts the structural type based on the neuronic two-layer feed-forward type neuron network of BP, utilizes Matlab6.5 Neural Network Toolbox function creation.Wherein input vector P dimension is 4, by [Ip, S, Tms, O
2] form; Hidden layer contains 5 neurons, adopts tansig as transfer function; Output layer contains 1 neuron, adopts the purelin transfer function.The network training function is got trainlm, adopts the newff function to generate network model.
BP network model creation procedure is as follows:
The %BP neuron network is used for the soft measurement of AFR
% loads sample data
load?ion_all；
sizeofp＝size(p)
sizeoft＝size(t)
% carries out standardization to sample data
[pn，meanp，stdp，tn，meant，stdt]＝prestd(p，t)；
% is divided into training set and test set with sample data
[R，Q]＝size(pn)
itst＝2：2：Q；
itr＝1：2：Q；
TestP=pn (:, itst); TestT=tn (:, itst); % test sample book collection
Ptr=pn (:, itr); Ttr=tn (:, itr); The % training sample set
% creates neuron network
net＝newff(minmax(ptr)，[51]，{′tansig＂purelin′}，′rainlm′)；
% trains network
net.trainParam.epochs＝500；
net.trainParam.goal＝0.001；
net＝init(net)；
[net，tr]＝train(net，ptr，ttr)
% carries out simulation analysis to network
an＝sim(net，testP)
y＝poststd(an，meant，stdt)；
% does linear regression analysis with simulation result and target output
figure；
[m，b，r]＝postreg(y，testT)；
end
Embodiment 1: use single oil spout mode during to the training of λ soft-sensing model and emulation, open the spray angle and be set in 10 ℃ of A of ATDC, after engine intake valve is opened, fuel oil will be entered cylinder by the intake valve of high temperature and the formed homogeneous charge of intake duct heating atomization and be evenly distributed on each position of cylinder, and promptly motor is in the homogeneous charge combustion regime.This moment stationary engine rotating speed 1800rpm, load 0.3MP gathers different air fuel ratio states 80 sample sets down between A/F=11～19 and divides and do two groups, one group is used as training sample set; Another group is as the simulation sample collection.Every group of sample set all is made of 40 input vectors and target vector.Wherein input vector has four elements, i.e. characteristic parameter and switching mode oxygen sensor signal such as the peak I p of spark plug ion current, integral value S and signal duration Tms, and target vector is an element with near the mixture strength spark plug (A/F).Element Ip, S and Tms detect with analytical system with spark plug ion current and obtain in the input vector, as shown in Figure 3; The switching mode zirconium oxide formula lambda sensor output that oxygen sensor signal is equipped with by former machine.The element of target vector is that mixture strength (A/F) records by linear air-fuel ratio detection system is actual near the spark plug, because when gathering sample is the homogeneous charge combustion regime, so the time cylinder that records in average mixture strength can think suitable with spark plug partial mixture concentration.
In the process of training, the weights of network and threshold value are adjusted repeatedly, to reduce the value of network performance function net.performFcn, up to the requirement that reaches setting.Because the Levenberg-Marquardt algorithm has the fastest convergence rate for medium-scale BP neuron network, we adopt the network training function trainlm of this algorithm that network is trained, and parameter is set:
net.trainParam.epochs＝500；net.trainParam.goal＝0.001；
In the hands-on, when network training is counted to 68 times, just reached the error requirements of setting, i.e. the mean square error mes=0.001 of network output and target output.Visible network has extraordinary learning performance.
After network training finishes, the output that we come artificial network with the simulation sample collection with the sim function, thus compare with target output, come the performance of supervising network.Simultaneously, our function postreg of providing with MATLAB does further analysis to the result of network training.Function postreg is a relation of utilizing the output of methods analyst network and the target output of linear regression, i.e. network output changes the variance ratio that changes with respect to target output, thus the training result of evaluating network.
Function postreg return of value R represents the correlation coefficient that network output and target are exported, and it is more near 1, and the output of expression network is approaching more with target output, and network performance is good more.By the figure shown in the function postreg, ideal regression straight line and optimum regression straight line almost overlap, and coefficient R=0.994 illustrates that the soft measurement network model of the λ that builds of institute has extraordinary mapping performance.
Claims (7)
1. spark plug partial mixture concentration flexible measurement method is characterized in that: may further comprise the steps:
A. obtain leading variable, set up mathematical model:
(1) rotating speed, load and the fuel injection quantity of extraction motor;
(2) gather ionic current, extract characteristic parameter Ip, S and Tms;
(3) pass through the parameter that the switching mode lambda sensor detects oxygen;
(4) set up the BP neural network soft sensor model;
B. obtain secondary variable, set up black-box model:
(1). obtain secondary variable Ip, S and Tms by the detection of spark plug ion current, oxygen sensor signal is obtained in the output of switching mode zirconium oxide formula lambda sensor, and linear air fuel ratio detects obtains A/F, filters the back and forms sample set;
(2). set up black-box model;
The error correction learning model carries out off-line training to the BP neural network soft sensor model, the soft-sensing model that trains is used for line measures.
2. spark plug partial mixture concentration flexible measurement method according to claim 1 is characterized in that: based on the BP neural network soft sensor model, create the λ SS neural network model towards Matlab, wherein input vector P dimension is 4, by [Ip, S, Tms, O
2] form; Hidden layer contains 5 neurons, adopts tansig as transfer function; Output layer contains 1 neuron, adopts the purelin transfer function; The network training function is got trainlm, adopts the newff function to generate network model.
3. spark plug partial mixture concentration flexible measurement method according to claim 1 and 2 is characterized in that: the step of setting up the BP neural network soft sensor model:
(1) loads sample data;
(2) sample data is carried out standardization;
(3) sample data is divided into training set and test set;
(4) create neuron network;
(5) network is trained;
(6) network is carried out simulation training;
(7) simulation result and target output are done linear regression analysis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2007101500368A CN101424219A (en) | 2007-11-01 | 2007-11-01 | Spark plug partial mixture concentration flexible measurement method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2007101500368A CN101424219A (en) | 2007-11-01 | 2007-11-01 | Spark plug partial mixture concentration flexible measurement method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101424219A true CN101424219A (en) | 2009-05-06 |
Family
ID=40615051
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2007101500368A Pending CN101424219A (en) | 2007-11-01 | 2007-11-01 | Spark plug partial mixture concentration flexible measurement method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101424219A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103748523A (en) * | 2011-08-22 | 2014-04-23 | 罗伯特·博世有限公司 | Method for the creation of a function for a control device |
| CN104656443A (en) * | 2014-12-31 | 2015-05-27 | 重庆邮电大学 | HCCI engine ignition timing self-adaptive PID control method based on BP neural network |
| CN105443259A (en) * | 2015-12-08 | 2016-03-30 | 上海海事大学 | Intelligent diesel engine cylinder balance adjusting algorithm based on approximate dynamic planning |
| CN108627357A (en) * | 2018-03-22 | 2018-10-09 | 西安科技大学 | A kind of coalcutter cutting load flexible measurement method |
| CN113239963A (en) * | 2021-04-13 | 2021-08-10 | 联合汽车电子有限公司 | Vehicle data processing method, device, equipment, vehicle and storage medium |
-
2007
- 2007-11-01 CN CNA2007101500368A patent/CN101424219A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103748523A (en) * | 2011-08-22 | 2014-04-23 | 罗伯特·博世有限公司 | Method for the creation of a function for a control device |
| CN105334746A (en) * | 2011-08-22 | 2016-02-17 | 罗伯特·博世有限公司 | Method for setting up functionality for control unit |
| CN103748523B (en) * | 2011-08-22 | 2016-12-28 | 罗伯特·博世有限公司 | For the method creating the function of control equipment |
| US9952567B2 (en) | 2011-08-22 | 2018-04-24 | Robert Bosch Gmbh | Method for setting up a functionality for a control unit |
| CN105334746B (en) * | 2011-08-22 | 2019-01-08 | 罗伯特·博世有限公司 | Method for creating the function of control equipment |
| CN104656443A (en) * | 2014-12-31 | 2015-05-27 | 重庆邮电大学 | HCCI engine ignition timing self-adaptive PID control method based on BP neural network |
| CN104656443B (en) * | 2014-12-31 | 2017-05-24 | 重庆邮电大学 | HCCI engine ignition timing self-adaptive PID control method based on BP neural network |
| CN105443259A (en) * | 2015-12-08 | 2016-03-30 | 上海海事大学 | Intelligent diesel engine cylinder balance adjusting algorithm based on approximate dynamic planning |
| CN105443259B (en) * | 2015-12-08 | 2017-12-01 | 上海海事大学 | A kind of intelligent regulation algorithm of cylinder of diesel engine balance based on approximate Dynamic Programming |
| CN108627357A (en) * | 2018-03-22 | 2018-10-09 | 西安科技大学 | A kind of coalcutter cutting load flexible measurement method |
| CN113239963A (en) * | 2021-04-13 | 2021-08-10 | 联合汽车电子有限公司 | Vehicle data processing method, device, equipment, vehicle and storage medium |
| CN113239963B (en) * | 2021-04-13 | 2024-03-01 | 联合汽车电子有限公司 | Method, device, equipment, vehicle and storage medium for processing vehicle data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Misfire detection of a turbocharged diesel engine by using artificial neural networks | |
| US20050251322A1 (en) | Virtual cylinder pressure sensor with individual estimators for pressure-related values | |
| CN101285427B (en) | Method for combined pulse spectrum controlling engine air admittance system | |
| CN104919298B (en) | Explosive motor based on exhaust manifold pressure misfires detection | |
| CN101424219A (en) | Spark plug partial mixture concentration flexible measurement method | |
| EP1705353A1 (en) | Method and device for estimating the inlet air flow in a combustion chamber of a cylinder of an internal combustion engine | |
| Tong et al. | Experiment analysis and computational optimization of the Atkinson cycle gasoline engine through NSGA Ⅱ algorithm using machine learning | |
| CN106762182A (en) | The control method and system of petrol engine transient detecting | |
| CN102567782A (en) | Neural-network-based automobile engine torque estimation method | |
| Bellone et al. | Comparison of CNN and LSTM for Modeling Virtual Sensors in an Engine | |
| Martínez-Morales et al. | Modeling engine fuel consumption and NOx with RBF neural network and MOPSO algorithm | |
| CN106321265A (en) | Method and system for identifying content of biodiesel in mixed fuel oil | |
| Guardiola et al. | Integration of intermittent measurement from in-cylinder pressure resonance in a multi-sensor mass flow estimator | |
| CN112883653B (en) | Artificial intelligence-based modeling method for real-time engine model | |
| CN113671564B (en) | NARX dynamic neural network-based microseism effective event automatic pickup method | |
| Khac et al. | Machine Learning Methods for Emissions Prediction in Combustion Engines with Multiple Cylinders | |
| Wang et al. | Exhaust pressure estimation and its application to variable geometry turbine and wastegate diagnostics | |
| Hafner et al. | Mechatronic design approach for engine management systems | |
| Taglialatela et al. | Real time prediction of particle sizing at the exhaust of a diesel engine by using a neural network model | |
| Cameretti et al. | Virtual calibration method for diesel engine by software in the loop techniques | |
| DE102017108995B3 (en) | Method and device for operating an internal combustion engine with a purging charge cycle | |
| Li et al. | Remote sensing and artificial neural network estimation of on-road vehicle emissions | |
| Brzozowski et al. | Toxicity of exhaust gases of compression ignition engine under conditions of variable load for different values of engine control parameters | |
| Ingesson et al. | An investigation on ignition-delay modelling for control | |
| Novella et al. | Identification of adequate combustion in turbulent jet ignition engines using machine learning algorithms |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Open date: 20090506 |